blob: 753ed9b5c70b941522108cee6ddeb7130b1f5d81 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000019#include "xfs_fs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
27#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_dinode.h"
30#include "xfs_inode.h"
Christoph Hellwigfd3200b2010-02-15 09:44:48 +000031#include "xfs_inode_item.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000032#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_error.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100034#include "xfs_vnodeops.h"
Christoph Hellwigf999a5b2008-11-28 14:23:32 +110035#include "xfs_da_btree.h"
Christoph Hellwigddcd8562008-12-03 07:55:34 -050036#include "xfs_ioctl.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000037#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010040#include <linux/falloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040042static const struct vm_operations_struct xfs_file_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Christoph Hellwigdda35b82010-02-15 09:44:46 +000044/*
Dave Chinner487f84f2011-01-12 11:37:10 +110045 * Locking primitives for read and write IO paths to ensure we consistently use
46 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
47 */
48static inline void
49xfs_rw_ilock(
50 struct xfs_inode *ip,
51 int type)
52{
53 if (type & XFS_IOLOCK_EXCL)
54 mutex_lock(&VFS_I(ip)->i_mutex);
55 xfs_ilock(ip, type);
56}
57
58static inline void
59xfs_rw_iunlock(
60 struct xfs_inode *ip,
61 int type)
62{
63 xfs_iunlock(ip, type);
64 if (type & XFS_IOLOCK_EXCL)
65 mutex_unlock(&VFS_I(ip)->i_mutex);
66}
67
68static inline void
69xfs_rw_ilock_demote(
70 struct xfs_inode *ip,
71 int type)
72{
73 xfs_ilock_demote(ip, type);
74 if (type & XFS_IOLOCK_EXCL)
75 mutex_unlock(&VFS_I(ip)->i_mutex);
76}
77
78/*
Christoph Hellwigdda35b82010-02-15 09:44:46 +000079 * xfs_iozero
80 *
81 * xfs_iozero clears the specified range of buffer supplied,
82 * and marks all the affected blocks as valid and modified. If
83 * an affected block is not allocated, it will be allocated. If
84 * an affected block is not completely overwritten, and is not
85 * valid before the operation, it will be read from disk before
86 * being partially zeroed.
87 */
88STATIC int
89xfs_iozero(
90 struct xfs_inode *ip, /* inode */
91 loff_t pos, /* offset in file */
92 size_t count) /* size of data to zero */
93{
94 struct page *page;
95 struct address_space *mapping;
96 int status;
97
98 mapping = VFS_I(ip)->i_mapping;
99 do {
100 unsigned offset, bytes;
101 void *fsdata;
102
103 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
104 bytes = PAGE_CACHE_SIZE - offset;
105 if (bytes > count)
106 bytes = count;
107
108 status = pagecache_write_begin(NULL, mapping, pos, bytes,
109 AOP_FLAG_UNINTERRUPTIBLE,
110 &page, &fsdata);
111 if (status)
112 break;
113
114 zero_user(page, offset, bytes);
115
116 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
117 page, fsdata);
118 WARN_ON(status <= 0); /* can't return less than zero! */
119 pos += bytes;
120 count -= bytes;
121 status = 0;
122 } while (count);
123
124 return (-status);
125}
126
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000127/*
128 * Fsync operations on directories are much simpler than on regular files,
129 * as there is no file data to flush, and thus also no need for explicit
130 * cache flush operations, and there are no non-transaction metadata updates
131 * on directories either.
132 */
133STATIC int
134xfs_dir_fsync(
135 struct file *file,
136 loff_t start,
137 loff_t end,
138 int datasync)
139{
140 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
141 struct xfs_mount *mp = ip->i_mount;
142 xfs_lsn_t lsn = 0;
143
144 trace_xfs_dir_fsync(ip);
145
146 xfs_ilock(ip, XFS_ILOCK_SHARED);
147 if (xfs_ipincount(ip))
148 lsn = ip->i_itemp->ili_last_lsn;
149 xfs_iunlock(ip, XFS_ILOCK_SHARED);
150
151 if (!lsn)
152 return 0;
153 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
154}
155
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000156STATIC int
157xfs_file_fsync(
158 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400159 loff_t start,
160 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000161 int datasync)
162{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200163 struct inode *inode = file->f_mapping->host;
164 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000165 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000166 struct xfs_trans *tp;
167 int error = 0;
168 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000169 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000170
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000171 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000172
Josef Bacik02c24a82011-07-16 20:44:56 -0400173 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
174 if (error)
175 return error;
176
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000177 if (XFS_FORCED_SHUTDOWN(mp))
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000178 return -XFS_ERROR(EIO);
179
180 xfs_iflags_clear(ip, XFS_ITRUNCATED);
181
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000182 if (mp->m_flags & XFS_MOUNT_BARRIER) {
183 /*
184 * If we have an RT and/or log subvolume we need to make sure
185 * to flush the write cache the device used for file data
186 * first. This is to ensure newly written file data make
187 * it to disk before logging the new inode size in case of
188 * an extending write.
189 */
190 if (XFS_IS_REALTIME_INODE(ip))
191 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
192 else if (mp->m_logdev_targp != mp->m_ddev_targp)
193 xfs_blkdev_issue_flush(mp->m_ddev_targp);
194 }
195
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000196 /*
197 * We always need to make sure that the required inode state is safe on
198 * disk. The inode might be clean but we still might need to force the
199 * log because of committed transactions that haven't hit the disk yet.
200 * Likewise, there could be unflushed non-transactional changes to the
201 * inode core that have to go to disk and this requires us to issue
202 * a synchronous transaction to capture these changes correctly.
203 *
204 * This code relies on the assumption that if the i_update_core field
205 * of the inode is clear and the inode is unpinned then it is clean
206 * and no action is required.
207 */
208 xfs_ilock(ip, XFS_ILOCK_SHARED);
209
Christoph Hellwig66d834e2010-02-15 09:44:49 +0000210 /*
211 * First check if the VFS inode is marked dirty. All the dirtying
212 * of non-transactional updates no goes through mark_inode_dirty*,
213 * which allows us to distinguish beteeen pure timestamp updates
214 * and i_size updates which need to be caught for fdatasync.
215 * After that also theck for the dirty state in the XFS inode, which
216 * might gets cleared when the inode gets written out via the AIL
217 * or xfs_iflush_cluster.
218 */
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200219 if (((inode->i_state & I_DIRTY_DATASYNC) ||
220 ((inode->i_state & I_DIRTY_SYNC) && !datasync)) &&
Christoph Hellwig66d834e2010-02-15 09:44:49 +0000221 ip->i_update_core) {
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000222 /*
223 * Kick off a transaction to log the inode core to get the
224 * updates. The sync transaction will also force the log.
225 */
226 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000227 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000228 error = xfs_trans_reserve(tp, 0,
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000229 XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000230 if (error) {
231 xfs_trans_cancel(tp, 0);
232 return -error;
233 }
234 xfs_ilock(ip, XFS_ILOCK_EXCL);
235
236 /*
237 * Note - it's possible that we might have pushed ourselves out
238 * of the way during trans_reserve which would flush the inode.
239 * But there's no guarantee that the inode buffer has actually
240 * gone out yet (it's delwri). Plus the buffer could be pinned
241 * anyway if it's part of an inode in another recent
242 * transaction. So we play it safe and fire off the
243 * transaction anyway.
244 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000245 xfs_trans_ijoin(tp, ip, 0);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000246 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Christoph Hellwigb1037052011-09-19 14:55:51 +0000247 error = xfs_trans_commit(tp, 0);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000248
Christoph Hellwigb1037052011-09-19 14:55:51 +0000249 lsn = ip->i_itemp->ili_last_lsn;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000250 xfs_iunlock(ip, XFS_ILOCK_EXCL);
251 } else {
252 /*
253 * Timestamps/size haven't changed since last inode flush or
254 * inode transaction commit. That means either nothing got
255 * written or a transaction committed which caught the updates.
256 * If the latter happened and the transaction hasn't hit the
257 * disk yet, the inode will be still be pinned. If it is,
258 * force the log.
259 */
Christoph Hellwigb1037052011-09-19 14:55:51 +0000260 if (xfs_ipincount(ip))
261 lsn = ip->i_itemp->ili_last_lsn;
Christoph Hellwig024910c2010-02-17 19:34:57 +0000262 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000263 }
264
Christoph Hellwigb1037052011-09-19 14:55:51 +0000265 if (!error && lsn)
266 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
267
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000268 /*
269 * If we only have a single device, and the log force about was
270 * a no-op we might have to flush the data device cache here.
271 * This can only happen for fdatasync/O_DSYNC if we were overwriting
272 * an already allocated file and thus do not have any metadata to
273 * commit.
274 */
275 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
276 mp->m_logdev_targp == mp->m_ddev_targp &&
277 !XFS_IS_REALTIME_INODE(ip) &&
278 !log_flushed)
279 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000280
281 return -error;
282}
283
Christoph Hellwig00258e32010-02-15 09:44:47 +0000284STATIC ssize_t
285xfs_file_aio_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000286 struct kiocb *iocb,
287 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000288 unsigned long nr_segs,
289 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000290{
291 struct file *file = iocb->ki_filp;
292 struct inode *inode = file->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000293 struct xfs_inode *ip = XFS_I(inode);
294 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000295 size_t size = 0;
296 ssize_t ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000297 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000298 xfs_fsize_t n;
299 unsigned long seg;
300
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000301 XFS_STATS_INC(xs_read_calls);
302
Christoph Hellwig00258e32010-02-15 09:44:47 +0000303 BUG_ON(iocb->ki_pos != pos);
304
305 if (unlikely(file->f_flags & O_DIRECT))
306 ioflags |= IO_ISDIRECT;
307 if (file->f_mode & FMODE_NOCMTIME)
308 ioflags |= IO_INVIS;
309
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000310 /* START copy & waste from filemap.c */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000311 for (seg = 0; seg < nr_segs; seg++) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000312 const struct iovec *iv = &iovp[seg];
313
314 /*
315 * If any segment has a negative length, or the cumulative
316 * length ever wraps negative then return -EINVAL.
317 */
318 size += iv->iov_len;
319 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
320 return XFS_ERROR(-EINVAL);
321 }
322 /* END copy & waste from filemap.c */
323
324 if (unlikely(ioflags & IO_ISDIRECT)) {
325 xfs_buftarg_t *target =
326 XFS_IS_REALTIME_INODE(ip) ?
327 mp->m_rtdev_targp : mp->m_ddev_targp;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000328 if ((iocb->ki_pos & target->bt_smask) ||
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000329 (size & target->bt_smask)) {
Christoph Hellwig00258e32010-02-15 09:44:47 +0000330 if (iocb->ki_pos == ip->i_size)
331 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000332 return -XFS_ERROR(EINVAL);
333 }
334 }
335
Christoph Hellwig00258e32010-02-15 09:44:47 +0000336 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
337 if (n <= 0 || size == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000338 return 0;
339
340 if (n < size)
341 size = n;
342
343 if (XFS_FORCED_SHUTDOWN(mp))
344 return -EIO;
345
Dave Chinner0c38a252011-08-25 07:17:01 +0000346 /*
347 * Locking is a bit tricky here. If we take an exclusive lock
348 * for direct IO, we effectively serialise all new concurrent
349 * read IO to this file and block it behind IO that is currently in
350 * progress because IO in progress holds the IO lock shared. We only
351 * need to hold the lock exclusive to blow away the page cache, so
352 * only take lock exclusively if the page cache needs invalidation.
353 * This allows the normal direct IO case of no page cache pages to
354 * proceeed concurrently without serialisation.
355 */
356 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
357 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
358 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100359 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
360
Christoph Hellwig00258e32010-02-15 09:44:47 +0000361 if (inode->i_mapping->nrpages) {
362 ret = -xfs_flushinval_pages(ip,
363 (iocb->ki_pos & PAGE_CACHE_MASK),
364 -1, FI_REMAPF_LOCKED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100365 if (ret) {
366 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
367 return ret;
368 }
Christoph Hellwig00258e32010-02-15 09:44:47 +0000369 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100370 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000371 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000372
Christoph Hellwig00258e32010-02-15 09:44:47 +0000373 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000374
Christoph Hellwig00258e32010-02-15 09:44:47 +0000375 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000376 if (ret > 0)
377 XFS_STATS_ADD(xs_read_bytes, ret);
378
Dave Chinner487f84f2011-01-12 11:37:10 +1100379 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000380 return ret;
381}
382
Christoph Hellwig00258e32010-02-15 09:44:47 +0000383STATIC ssize_t
384xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000385 struct file *infilp,
386 loff_t *ppos,
387 struct pipe_inode_info *pipe,
388 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000389 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000390{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000391 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000392 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000393 ssize_t ret;
394
395 XFS_STATS_INC(xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000396
397 if (infilp->f_mode & FMODE_NOCMTIME)
398 ioflags |= IO_INVIS;
399
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000400 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
401 return -EIO;
402
Dave Chinner487f84f2011-01-12 11:37:10 +1100403 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000404
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000405 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
406
407 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
408 if (ret > 0)
409 XFS_STATS_ADD(xs_read_bytes, ret);
410
Dave Chinner487f84f2011-01-12 11:37:10 +1100411 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000412 return ret;
413}
414
Dave Chinneredafb6d2011-01-11 10:14:06 +1100415STATIC void
416xfs_aio_write_isize_update(
417 struct inode *inode,
418 loff_t *ppos,
419 ssize_t bytes_written)
420{
421 struct xfs_inode *ip = XFS_I(inode);
422 xfs_fsize_t isize = i_size_read(inode);
423
424 if (bytes_written > 0)
425 XFS_STATS_ADD(xs_write_bytes, bytes_written);
426
427 if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
428 *ppos > isize))
429 *ppos = isize;
430
431 if (*ppos > ip->i_size) {
Dave Chinner487f84f2011-01-12 11:37:10 +1100432 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinneredafb6d2011-01-11 10:14:06 +1100433 if (*ppos > ip->i_size)
434 ip->i_size = *ppos;
Dave Chinner487f84f2011-01-12 11:37:10 +1100435 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinneredafb6d2011-01-11 10:14:06 +1100436 }
437}
438
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100439/*
440 * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300441 * part of the I/O may have been written to disk before the error occurred. In
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100442 * this case the on-disk file size may have been adjusted beyond the in-memory
443 * file size and now needs to be truncated back.
444 */
445STATIC void
446xfs_aio_write_newsize_update(
Dave Chinner7271d242011-08-25 07:17:02 +0000447 struct xfs_inode *ip,
448 xfs_fsize_t new_size)
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100449{
Dave Chinner7271d242011-08-25 07:17:02 +0000450 if (new_size == ip->i_new_size) {
Dave Chinner487f84f2011-01-12 11:37:10 +1100451 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner7271d242011-08-25 07:17:02 +0000452 if (new_size == ip->i_new_size)
453 ip->i_new_size = 0;
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100454 if (ip->i_d.di_size > ip->i_size)
455 ip->i_d.di_size = ip->i_size;
Dave Chinner487f84f2011-01-12 11:37:10 +1100456 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100457 }
458}
459
Dave Chinner487f84f2011-01-12 11:37:10 +1100460/*
461 * xfs_file_splice_write() does not use xfs_rw_ilock() because
462 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
463 * couuld cause lock inversions between the aio_write path and the splice path
464 * if someone is doing concurrent splice(2) based writes and write(2) based
465 * writes to the same inode. The only real way to fix this is to re-implement
466 * the generic code here with correct locking orders.
467 */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000468STATIC ssize_t
469xfs_file_splice_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000470 struct pipe_inode_info *pipe,
471 struct file *outfilp,
472 loff_t *ppos,
473 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000474 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000475{
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000476 struct inode *inode = outfilp->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000477 struct xfs_inode *ip = XFS_I(inode);
Dave Chinneredafb6d2011-01-11 10:14:06 +1100478 xfs_fsize_t new_size;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000479 int ioflags = 0;
480 ssize_t ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000481
482 XFS_STATS_INC(xs_write_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000483
484 if (outfilp->f_mode & FMODE_NOCMTIME)
485 ioflags |= IO_INVIS;
486
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000487 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
488 return -EIO;
489
490 xfs_ilock(ip, XFS_IOLOCK_EXCL);
491
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000492 new_size = *ppos + count;
493
494 xfs_ilock(ip, XFS_ILOCK_EXCL);
495 if (new_size > ip->i_size)
496 ip->i_new_size = new_size;
497 xfs_iunlock(ip, XFS_ILOCK_EXCL);
498
499 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
500
501 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000502
Dave Chinneredafb6d2011-01-11 10:14:06 +1100503 xfs_aio_write_isize_update(inode, ppos, ret);
Dave Chinner7271d242011-08-25 07:17:02 +0000504 xfs_aio_write_newsize_update(ip, new_size);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000505 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
506 return ret;
507}
508
509/*
510 * This routine is called to handle zeroing any space in the last
511 * block of the file that is beyond the EOF. We do this since the
512 * size is being increased without writing anything to that block
513 * and we don't want anyone to read the garbage on the disk.
514 */
515STATIC int /* error (positive) */
516xfs_zero_last_block(
517 xfs_inode_t *ip,
518 xfs_fsize_t offset,
519 xfs_fsize_t isize)
520{
521 xfs_fileoff_t last_fsb;
522 xfs_mount_t *mp = ip->i_mount;
523 int nimaps;
524 int zero_offset;
525 int zero_len;
526 int error = 0;
527 xfs_bmbt_irec_t imap;
528
529 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
530
531 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
532 if (zero_offset == 0) {
533 /*
534 * There are no extra bytes in the last block on disk to
535 * zero, so return.
536 */
537 return 0;
538 }
539
540 last_fsb = XFS_B_TO_FSBT(mp, isize);
541 nimaps = 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000542 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
543 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000544 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000545 ASSERT(nimaps > 0);
546 /*
547 * If the block underlying isize is just a hole, then there
548 * is nothing to zero.
549 */
550 if (imap.br_startblock == HOLESTARTBLOCK) {
551 return 0;
552 }
553 /*
554 * Zero the part of the last block beyond the EOF, and write it
555 * out sync. We need to drop the ilock while we do this so we
556 * don't deadlock when the buffer cache calls back to us.
557 */
558 xfs_iunlock(ip, XFS_ILOCK_EXCL);
559
560 zero_len = mp->m_sb.sb_blocksize - zero_offset;
561 if (isize + zero_len > offset)
562 zero_len = offset - isize;
563 error = xfs_iozero(ip, isize, zero_len);
564
565 xfs_ilock(ip, XFS_ILOCK_EXCL);
566 ASSERT(error >= 0);
567 return error;
568}
569
570/*
571 * Zero any on disk space between the current EOF and the new,
572 * larger EOF. This handles the normal case of zeroing the remainder
573 * of the last block in the file and the unusual case of zeroing blocks
574 * out beyond the size of the file. This second case only happens
575 * with fixed size extents and when the system crashes before the inode
576 * size was updated but after blocks were allocated. If fill is set,
577 * then any holes in the range are filled and zeroed. If not, the holes
578 * are left alone as holes.
579 */
580
581int /* error (positive) */
582xfs_zero_eof(
583 xfs_inode_t *ip,
584 xfs_off_t offset, /* starting I/O offset */
585 xfs_fsize_t isize) /* current inode size */
586{
587 xfs_mount_t *mp = ip->i_mount;
588 xfs_fileoff_t start_zero_fsb;
589 xfs_fileoff_t end_zero_fsb;
590 xfs_fileoff_t zero_count_fsb;
591 xfs_fileoff_t last_fsb;
592 xfs_fileoff_t zero_off;
593 xfs_fsize_t zero_len;
594 int nimaps;
595 int error = 0;
596 xfs_bmbt_irec_t imap;
597
598 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
599 ASSERT(offset > isize);
600
601 /*
602 * First handle zeroing the block on which isize resides.
603 * We only zero a part of that block so it is handled specially.
604 */
605 error = xfs_zero_last_block(ip, offset, isize);
606 if (error) {
607 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
608 return error;
609 }
610
611 /*
612 * Calculate the range between the new size and the old
613 * where blocks needing to be zeroed may exist. To get the
614 * block where the last byte in the file currently resides,
615 * we need to subtract one from the size and truncate back
616 * to a block boundary. We subtract 1 in case the size is
617 * exactly on a block boundary.
618 */
619 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
620 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
621 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
622 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
623 if (last_fsb == end_zero_fsb) {
624 /*
625 * The size was only incremented on its last block.
626 * We took care of that above, so just return.
627 */
628 return 0;
629 }
630
631 ASSERT(start_zero_fsb <= end_zero_fsb);
632 while (start_zero_fsb <= end_zero_fsb) {
633 nimaps = 1;
634 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000635 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
636 &imap, &nimaps, 0);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000637 if (error) {
638 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
639 return error;
640 }
641 ASSERT(nimaps > 0);
642
643 if (imap.br_state == XFS_EXT_UNWRITTEN ||
644 imap.br_startblock == HOLESTARTBLOCK) {
645 /*
646 * This loop handles initializing pages that were
647 * partially initialized by the code below this
648 * loop. It basically zeroes the part of the page
649 * that sits on a hole and sets the page as P_HOLE
650 * and calls remapf if it is a mapped file.
651 */
652 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
653 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
654 continue;
655 }
656
657 /*
658 * There are blocks we need to zero.
659 * Drop the inode lock while we're doing the I/O.
660 * We'll still have the iolock to protect us.
661 */
662 xfs_iunlock(ip, XFS_ILOCK_EXCL);
663
664 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
665 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
666
667 if ((zero_off + zero_len) > offset)
668 zero_len = offset - zero_off;
669
670 error = xfs_iozero(ip, zero_off, zero_len);
671 if (error) {
672 goto out_lock;
673 }
674
675 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
676 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
677
678 xfs_ilock(ip, XFS_ILOCK_EXCL);
679 }
680
681 return 0;
682
683out_lock:
684 xfs_ilock(ip, XFS_ILOCK_EXCL);
685 ASSERT(error >= 0);
686 return error;
687}
688
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100689/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100690 * Common pre-write limit and setup checks.
691 *
692 * Returns with iolock held according to @iolock.
693 */
694STATIC ssize_t
695xfs_file_aio_write_checks(
696 struct file *file,
697 loff_t *pos,
698 size_t *count,
Dave Chinner7271d242011-08-25 07:17:02 +0000699 xfs_fsize_t *new_sizep,
Dave Chinner4d8d1582011-01-11 10:23:42 +1100700 int *iolock)
701{
702 struct inode *inode = file->f_mapping->host;
703 struct xfs_inode *ip = XFS_I(inode);
704 xfs_fsize_t new_size;
705 int error = 0;
706
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000707 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner7271d242011-08-25 07:17:02 +0000708 *new_sizep = 0;
709restart:
Dave Chinner4d8d1582011-01-11 10:23:42 +1100710 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
711 if (error) {
712 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
713 *iolock = 0;
714 return error;
715 }
716
Dave Chinner4d8d1582011-01-11 10:23:42 +1100717 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
718 file_update_time(file);
719
720 /*
721 * If the offset is beyond the size of the file, we need to zero any
722 * blocks that fall between the existing EOF and the start of this
Dave Chinner7271d242011-08-25 07:17:02 +0000723 * write. There is no need to issue zeroing if another in-flght IO ends
724 * at or before this one If zeronig is needed and we are currently
725 * holding the iolock shared, we need to update it to exclusive which
726 * involves dropping all locks and relocking to maintain correct locking
727 * order. If we do this, restart the function to ensure all checks and
728 * values are still valid.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100729 */
Dave Chinner7271d242011-08-25 07:17:02 +0000730 if ((ip->i_new_size && *pos > ip->i_new_size) ||
731 (!ip->i_new_size && *pos > ip->i_size)) {
732 if (*iolock == XFS_IOLOCK_SHARED) {
733 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
734 *iolock = XFS_IOLOCK_EXCL;
735 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
736 goto restart;
737 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100738 error = -xfs_zero_eof(ip, *pos, ip->i_size);
Dave Chinner7271d242011-08-25 07:17:02 +0000739 }
740
741 /*
742 * If this IO extends beyond EOF, we may need to update ip->i_new_size.
743 * We have already zeroed space beyond EOF (if necessary). Only update
744 * ip->i_new_size if this IO ends beyond any other in-flight writes.
745 */
746 new_size = *pos + *count;
747 if (new_size > ip->i_size) {
748 if (new_size > ip->i_new_size)
749 ip->i_new_size = new_size;
750 *new_sizep = new_size;
751 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100752
753 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
754 if (error)
755 return error;
756
757 /*
758 * If we're writing the file then make sure to clear the setuid and
759 * setgid bits if the process is not being run by root. This keeps
760 * people from modifying setuid and setgid binaries.
761 */
762 return file_remove_suid(file);
763
764}
765
766/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100767 * xfs_file_dio_aio_write - handle direct IO writes
768 *
769 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100770 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100771 * follow locking changes and looping.
772 *
Dave Chinnereda77982011-01-11 10:22:40 +1100773 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
774 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
775 * pages are flushed out.
776 *
777 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
778 * allowing them to be done in parallel with reads and other direct IO writes.
779 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
780 * needs to do sub-block zeroing and that requires serialisation against other
781 * direct IOs to the same block. In this case we need to serialise the
782 * submission of the unaligned IOs so that we don't get racing block zeroing in
783 * the dio layer. To avoid the problem with aio, we also need to wait for
784 * outstanding IOs to complete so that unwritten extent conversion is completed
785 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000786 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100787 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100788 * Returns with locks held indicated by @iolock and errors indicated by
789 * negative return values.
790 */
791STATIC ssize_t
792xfs_file_dio_aio_write(
793 struct kiocb *iocb,
794 const struct iovec *iovp,
795 unsigned long nr_segs,
796 loff_t pos,
797 size_t ocount,
Dave Chinner7271d242011-08-25 07:17:02 +0000798 xfs_fsize_t *new_size,
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100799 int *iolock)
800{
801 struct file *file = iocb->ki_filp;
802 struct address_space *mapping = file->f_mapping;
803 struct inode *inode = mapping->host;
804 struct xfs_inode *ip = XFS_I(inode);
805 struct xfs_mount *mp = ip->i_mount;
806 ssize_t ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100807 size_t count = ocount;
Dave Chinnereda77982011-01-11 10:22:40 +1100808 int unaligned_io = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100809 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
810 mp->m_rtdev_targp : mp->m_ddev_targp;
811
812 *iolock = 0;
813 if ((pos & target->bt_smask) || (count & target->bt_smask))
814 return -XFS_ERROR(EINVAL);
815
Dave Chinnereda77982011-01-11 10:22:40 +1100816 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
817 unaligned_io = 1;
818
Dave Chinner7271d242011-08-25 07:17:02 +0000819 /*
820 * We don't need to take an exclusive lock unless there page cache needs
821 * to be invalidated or unaligned IO is being executed. We don't need to
822 * consider the EOF extension case here because
823 * xfs_file_aio_write_checks() will relock the inode as necessary for
824 * EOF zeroing cases and fill out the new inode size as appropriate.
825 */
826 if (unaligned_io || mapping->nrpages)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100827 *iolock = XFS_IOLOCK_EXCL;
828 else
829 *iolock = XFS_IOLOCK_SHARED;
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000830 xfs_rw_ilock(ip, *iolock);
831
832 /*
833 * Recheck if there are cached pages that need invalidate after we got
834 * the iolock to protect against other threads adding new pages while
835 * we were waiting for the iolock.
836 */
837 if (mapping->nrpages && *iolock == XFS_IOLOCK_SHARED) {
838 xfs_rw_iunlock(ip, *iolock);
839 *iolock = XFS_IOLOCK_EXCL;
840 xfs_rw_ilock(ip, *iolock);
841 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100842
Dave Chinner7271d242011-08-25 07:17:02 +0000843 ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100844 if (ret)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100845 return ret;
846
847 if (mapping->nrpages) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100848 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
849 FI_REMAPF_LOCKED);
850 if (ret)
851 return ret;
852 }
853
Dave Chinnereda77982011-01-11 10:22:40 +1100854 /*
855 * If we are doing unaligned IO, wait for all other IO to drain,
856 * otherwise demote the lock if we had to flush cached pages
857 */
858 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000859 inode_dio_wait(inode);
Dave Chinnereda77982011-01-11 10:22:40 +1100860 else if (*iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100861 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
862 *iolock = XFS_IOLOCK_SHARED;
863 }
864
865 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
866 ret = generic_file_direct_write(iocb, iovp,
867 &nr_segs, pos, &iocb->ki_pos, count, ocount);
868
869 /* No fallback to buffered IO on errors for XFS. */
870 ASSERT(ret < 0 || ret == count);
871 return ret;
872}
873
Christoph Hellwig00258e32010-02-15 09:44:47 +0000874STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100875xfs_file_buffered_aio_write(
876 struct kiocb *iocb,
877 const struct iovec *iovp,
878 unsigned long nr_segs,
879 loff_t pos,
880 size_t ocount,
Dave Chinner7271d242011-08-25 07:17:02 +0000881 xfs_fsize_t *new_size,
Dave Chinner637bbc72011-01-11 10:17:30 +1100882 int *iolock)
883{
884 struct file *file = iocb->ki_filp;
885 struct address_space *mapping = file->f_mapping;
886 struct inode *inode = mapping->host;
887 struct xfs_inode *ip = XFS_I(inode);
888 ssize_t ret;
889 int enospc = 0;
Dave Chinner637bbc72011-01-11 10:17:30 +1100890 size_t count = ocount;
891
892 *iolock = XFS_IOLOCK_EXCL;
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000893 xfs_rw_ilock(ip, *iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100894
Dave Chinner7271d242011-08-25 07:17:02 +0000895 ret = xfs_file_aio_write_checks(file, &pos, &count, new_size, iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100896 if (ret)
Dave Chinner637bbc72011-01-11 10:17:30 +1100897 return ret;
898
899 /* We can write back this queue in page reclaim */
900 current->backing_dev_info = mapping->backing_dev_info;
901
902write_retry:
903 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
904 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
905 pos, &iocb->ki_pos, count, ret);
906 /*
907 * if we just got an ENOSPC, flush the inode now we aren't holding any
908 * page locks and retry *once*
909 */
910 if (ret == -ENOSPC && !enospc) {
911 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
912 if (ret)
913 return ret;
914 enospc = 1;
915 goto write_retry;
916 }
917 current->backing_dev_info = NULL;
918 return ret;
919}
920
921STATIC ssize_t
Christoph Hellwig00258e32010-02-15 09:44:47 +0000922xfs_file_aio_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000923 struct kiocb *iocb,
924 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000925 unsigned long nr_segs,
926 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000927{
928 struct file *file = iocb->ki_filp;
929 struct address_space *mapping = file->f_mapping;
930 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000931 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100932 ssize_t ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000933 int iolock;
Dave Chinner637bbc72011-01-11 10:17:30 +1100934 size_t ocount = 0;
Dave Chinner7271d242011-08-25 07:17:02 +0000935 xfs_fsize_t new_size = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000936
937 XFS_STATS_INC(xs_write_calls);
938
Christoph Hellwig00258e32010-02-15 09:44:47 +0000939 BUG_ON(iocb->ki_pos != pos);
940
Dave Chinnera363f0c2011-01-11 10:13:53 +1100941 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
942 if (ret)
943 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000944
Dave Chinner637bbc72011-01-11 10:17:30 +1100945 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000946 return 0;
947
Dave Chinner637bbc72011-01-11 10:17:30 +1100948 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000949
Dave Chinner637bbc72011-01-11 10:17:30 +1100950 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000951 return -EIO;
952
Dave Chinner637bbc72011-01-11 10:17:30 +1100953 if (unlikely(file->f_flags & O_DIRECT))
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100954 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos,
Dave Chinner7271d242011-08-25 07:17:02 +0000955 ocount, &new_size, &iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100956 else
957 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
Dave Chinner7271d242011-08-25 07:17:02 +0000958 ocount, &new_size, &iolock);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000959
Dave Chinneredafb6d2011-01-11 10:14:06 +1100960 xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000961
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000962 if (ret <= 0)
Dave Chinner637bbc72011-01-11 10:17:30 +1100963 goto out_unlock;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000964
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000965 /* Handle various SYNC-type writes */
966 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
967 loff_t end = pos + ret - 1;
Markus Trippelsdorf340a0a02011-07-24 14:03:30 +0200968 int error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000969
Dave Chinner487f84f2011-01-12 11:37:10 +1100970 xfs_rw_iunlock(ip, iolock);
Markus Trippelsdorf340a0a02011-07-24 14:03:30 +0200971 error = xfs_file_fsync(file, pos, end,
Josef Bacik02c24a82011-07-16 20:44:56 -0400972 (file->f_flags & __O_SYNC) ? 0 : 1);
Dave Chinner487f84f2011-01-12 11:37:10 +1100973 xfs_rw_ilock(ip, iolock);
Markus Trippelsdorf340a0a02011-07-24 14:03:30 +0200974 if (error)
975 ret = error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000976 }
977
Dave Chinner637bbc72011-01-11 10:17:30 +1100978out_unlock:
Dave Chinner7271d242011-08-25 07:17:02 +0000979 xfs_aio_write_newsize_update(ip, new_size);
Dave Chinner487f84f2011-01-12 11:37:10 +1100980 xfs_rw_iunlock(ip, iolock);
Dave Chinnera363f0c2011-01-11 10:13:53 +1100981 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000982}
983
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100984STATIC long
985xfs_file_fallocate(
986 struct file *file,
987 int mode,
988 loff_t offset,
989 loff_t len)
990{
991 struct inode *inode = file->f_path.dentry->d_inode;
992 long error;
993 loff_t new_size = 0;
994 xfs_flock64_t bf;
995 xfs_inode_t *ip = XFS_I(inode);
996 int cmd = XFS_IOC_RESVSP;
Dave Chinner82878892011-03-26 09:13:08 +1100997 int attr_flags = XFS_ATTR_NOLOCK;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100998
999 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1000 return -EOPNOTSUPP;
1001
1002 bf.l_whence = 0;
1003 bf.l_start = offset;
1004 bf.l_len = len;
1005
1006 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1007
1008 if (mode & FALLOC_FL_PUNCH_HOLE)
1009 cmd = XFS_IOC_UNRESVSP;
1010
1011 /* check the new inode size is valid before allocating */
1012 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1013 offset + len > i_size_read(inode)) {
1014 new_size = offset + len;
1015 error = inode_newsize_ok(inode, new_size);
1016 if (error)
1017 goto out_unlock;
1018 }
1019
Dave Chinner82878892011-03-26 09:13:08 +11001020 if (file->f_flags & O_DSYNC)
1021 attr_flags |= XFS_ATTR_SYNC;
1022
1023 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001024 if (error)
1025 goto out_unlock;
1026
1027 /* Change file size if needed */
1028 if (new_size) {
1029 struct iattr iattr;
1030
1031 iattr.ia_valid = ATTR_SIZE;
1032 iattr.ia_size = new_size;
Christoph Hellwigc4ed4242011-07-08 14:34:23 +02001033 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001034 }
1035
1036out_unlock:
1037 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1038 return error;
1039}
1040
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001043xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001045 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001047 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001049 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1050 return -EIO;
1051 return 0;
1052}
1053
1054STATIC int
1055xfs_dir_open(
1056 struct inode *inode,
1057 struct file *file)
1058{
1059 struct xfs_inode *ip = XFS_I(inode);
1060 int mode;
1061 int error;
1062
1063 error = xfs_file_open(inode, file);
1064 if (error)
1065 return error;
1066
1067 /*
1068 * If there are any blocks, read-ahead block 0 as we're almost
1069 * certain to have the next operation be a read there.
1070 */
1071 mode = xfs_ilock_map_shared(ip);
1072 if (ip->i_d.di_nextents > 0)
1073 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
1074 xfs_iunlock(ip, mode);
1075 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076}
1077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001079xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 struct inode *inode,
1081 struct file *filp)
1082{
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001083 return -xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084}
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001087xfs_file_readdir(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 struct file *filp,
1089 void *dirent,
1090 filldir_t filldir)
1091{
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001092 struct inode *inode = filp->f_path.dentry->d_inode;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001093 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001094 int error;
1095 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001097 /*
1098 * The Linux API doesn't pass down the total size of the buffer
1099 * we read into down to the filesystem. With the filldir concept
1100 * it's not needed for correct information, but the XFS dir2 leaf
1101 * code wants an estimate of the buffer size to calculate it's
1102 * readahead window and size the buffers used for mapping to
1103 * physical blocks.
1104 *
1105 * Try to give it an estimate that's good enough, maybe at some
1106 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +00001107 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001108 */
Eric Sandeena9cc7992010-02-03 17:50:13 +00001109 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001111 error = xfs_readdir(ip, dirent, bufsize,
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001112 (xfs_off_t *)&filp->f_pos, filldir);
1113 if (error)
1114 return -error;
1115 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116}
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001119xfs_file_mmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 struct file *filp,
1121 struct vm_area_struct *vma)
1122{
Nathan Scott3562fd42006-03-14 14:00:35 +11001123 vma->vm_ops = &xfs_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001124 vma->vm_flags |= VM_CAN_NONLINEAR;
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001125
Nathan Scottfbc14622006-06-09 14:52:13 +10001126 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 return 0;
1128}
1129
David Chinner4f57dbc2007-07-19 16:28:17 +10001130/*
1131 * mmap()d file has taken write protection fault and is being made
1132 * writable. We can set the page state up correctly for a writable
1133 * page, which means we can do correct delalloc accounting (ENOSPC
1134 * checking!) and unwritten extent mapping.
1135 */
1136STATIC int
1137xfs_vm_page_mkwrite(
1138 struct vm_area_struct *vma,
Nick Pigginc2ec1752009-03-31 15:23:21 -07001139 struct vm_fault *vmf)
David Chinner4f57dbc2007-07-19 16:28:17 +10001140{
Nick Pigginc2ec1752009-03-31 15:23:21 -07001141 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
David Chinner4f57dbc2007-07-19 16:28:17 +10001142}
1143
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001144const struct file_operations xfs_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 .llseek = generic_file_llseek,
1146 .read = do_sync_read,
Dean Roehrichbb3f7242005-09-02 15:43:05 +10001147 .write = do_sync_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001148 .aio_read = xfs_file_aio_read,
1149 .aio_write = xfs_file_aio_write,
Nathan Scott1b895842006-03-31 13:08:59 +10001150 .splice_read = xfs_file_splice_read,
1151 .splice_write = xfs_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001152 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001154 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001156 .mmap = xfs_file_mmap,
1157 .open = xfs_file_open,
1158 .release = xfs_file_release,
1159 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001160 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161};
1162
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001163const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001164 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 .read = generic_read_dir,
Nathan Scott3562fd42006-03-14 14:00:35 +11001166 .readdir = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001167 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001168 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001169#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001170 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001171#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001172 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173};
1174
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001175static const struct vm_operations_struct xfs_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001176 .fault = filemap_fault,
David Chinner4f57dbc2007-07-19 16:28:17 +10001177 .page_mkwrite = xfs_vm_page_mkwrite,
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001178};